Research map / Label-efficient and robust learning / Test-time adaptation
Distribution Alignment for Fully Test-Time Adaptation with Dynamic Online Data Streams
Ziqiang Wang, Zhixiang Chi, Yanan Wu, Li Gu, Zhi Liu, Konstantinos N Plataniotis, Yang Wang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| align test feature distributions to source statistics (DA loss) instead of entropy-based self-training | TTA with BN affine layers · CIFAR10-C non-i.i.d. | mean error (%): 27.6 (RoTTA) -> 24.3 | Improved |
| apply DA-TTA on CIFAR100-C non-i.i.d. | TTA · CIFAR100-C non-i.i.d. | mean error (%): 39.3 (MEMO) -> 31.6 | Improved |
| adapt with TTBN or entropy-based self-training under non-i.i.d. streams | TTBN, TENT, SAR, EATA, CoTTA · CIFAR10-C non-i.i.d. | mean error (%): 43.5 (source) -> 75.7 (TTBN), 77.0 (TENT) | Hurt |
| add entropy minimization loss on top of DA loss (ablation) | DA-TTA · ImageNet-C non-i.i.d. | mean error (%): 70.1 -> 64.8 | Improved |
| add entropy minimization loss on TTBN baseline (TENT) (ablation) | TTBN · CIFAR10-C non-i.i.d. | mean error (%): 75.7 -> 77.0 | Hurt |
| apply DA loss to all layers instead of only the latter half (ablation) | DA-TTA · ImageNet-C non-i.i.d. | mean error (%): 68.1 -> 64.8 | Improved |
| apply DA loss to all layers instead of only the former half (ablation) | DA-TTA · CIFAR10-C non-i.i.d. | mean error (%): 24.9 -> 24.3 | Improved |
| add domain shift detection with affine layer reset in continual TTA | DA-TTA · ImageNet-C continual non-i.i.d. | mean error (%): 70.7 (LAME) -> 68.6; most competitors above 90% error | Improved |
| use DA-TTA on realistic domain shifts versus source model | DA-TTA · ImageNet-R non-i.i.d. | error (%): 63.8 -> 58.5 | Improved |
| use DA-TTA on adversarial ImageNet-A | DA-TTA · ImageNet-A non-i.i.d. | error (%): 90.5 -> 88.6 | Improved |
| reduce batch size or increase label shift severity (ablation) | existing TTA methods · non-i.i.d. streams | error | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- TENT — Tent: Fully Test-Time Adaptation by Entropy Minimization (ICLR 2021) · better on 2 of 2 shared results
- NOTE — NOTE: Robust Continual Test-time Adaptation Against Temporal Correlation (NeurIPS 2022) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- A Layer Selection Approach to Test Time Adaptation (AAAI 2025 · similarity 0.91)
- Towards Real-World Test-Time Adaptation: Tri-net Self-Training with Balanced Normalization (AAAI 2024 · similarity 0.91)
- Active Test-Time Adaptation: Theoretical Analyses and An Algorithm (ICLR 2024 · similarity 0.90)
- AETTA: Label-Free Accuracy Estimation for Test-Time Adaptation (CVPR 2024 · similarity 0.90)
- Test-time Correlation Alignment (ICML 2025 · similarity 0.90)
- Evaluation of Test-Time Adaptation Under Computational Time Constraints (ICML 2024 · similarity 0.90)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
